SovereignPA-Bench Evaluates User-Owned Personal Agents Under Consent Constraints

arXiv·medium signal

A benchmark for persistent, user-owned personal agents that remember preferences, filter platform-mediated information, and negotiate with services under evolving intent and consent constraints — dimensions existing tool-use and web-navigation benchmarks miss. It gives personal-agent builders an evaluation scaffold beyond task completion. Timely as 'sovereign' user-side agents move from concept to product.